Background The findings from prospective studies on visual impairment (VI) and the risk of mortality are not consistent.
Objective A meta-analysis of prospective studies was conducted to quantitatively summarise the evidence about the association between VI and the risk of mortality.
Methods Pertinent studies were identified by a search of PubMed, Web of Science and the Chinese National Knowledge Infrastructure and Wanfang databases up to December 2015. The random-effect model was used to combine study-specific relative risks (RRs) and 95% CIs. Meta-regression and subgroup analysis were conducted to explore potential sources of heterogeneity. Publication bias was estimated by Egger's test and the funnel plot. Dose–response relationship was assessed by restricted cubic spline functions.
Results This meta-analysis contained 29 prospective studies including 269 839 participants and 67 061 deaths. Compared to the no VI, the highest VI level was significantly associated with an increased risk of mortality (RR: 1.36, 95% CI 1.25 to 1.46). The association remained significant in participants older than 65 years (RR: 1.28, 95% CI 1.18 to 1.39), and a significant association was also observed in men (RR: 1.29, 95% CI 1.07 to 1.54) and women (RR: 1.39, 95% CI 1.14 to 1.70), respectively. For dose–response analysis, a linear relation was found between visual acuity (VA) and the risk of mortality. For every 0.1 Logarithm of the Minimum Angle of Resolution (LogMAR) increment, the risk of mortality increased by 4% (RR: 1.04, 95% CI 1.01 to 1.06).
Conclusions VI was significantly associated with an increased risk of mortality.
- META ANALYSIS
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
According to the WHO, the latest definition of visual impairment (VI) is presenting vision worse than 6/18. Blindness is defined as either ‘irreversible’ blindness (no perception of light) or light perception in one eye with vision worse than 3/60 in the better eye.1 The major causes include uncorrected refractive error (43%), unoperated cataract (33%) and glaucoma (2%).2 On the basis of this definition, there were 285 million persons of all ages with VI in 2010 worldwide, of whom 39 million were blind3 and about 65% are aged 50 and older.2 With the ageing of the population, VI is becoming a significant public health problem.4
Many studies found that VI impacted people's health by increasing the risks of falls,5 ,6 fractures7 ,8 and accidents.9 In addition, VI impacts negatively on functional independence. It increases the need for community services10 and institutionalisation.11 ,12 The reason is that the VI may decreases the ability to perform basic and instrumental activities of daily living.13 ,14 All the adverse health effects may increase the risk of mortality. For example, a previous study reported significant associations between bone fracture and the risk of mortality in general populations.15 On account of the above associations, many epidemiological studies have explored the association between VI and the risk of mortality. However, the association is conflicting in observational studies.4 ,16–37 VI was reported to be associated with the risk of mortality in several studies,4 ,16–20 ,22 ,23 ,25 ,26 ,28 ,30 ,32 ,34 ,35 ,37 whereas no association was found in other studies.21 ,24 ,27–29 ,31 ,33 ,35 ,36 Furthermore, the findings were inconsistent when men and women were analysed separately as well as in participants older than 65 years, and the dose–response relationship between visual acuity (VA) and the risk of mortality has not been well defined.
For those reasons, the meta-analysis was performed by combining all available information of prospective studies to assess the association of VI and the risk of mortality in the general population, in participants older than 65 years and in men and women, respectively.
Literature search strategy
To identify all available English and Chinese-related articles, a literature search (up to December 2015) was performed in PubMed, Web of Science and the Chinese National Knowledge Infrastructure and Wanfang databases with the following keywords: ‘visual impairment’ (or ‘visual acuity’ or ‘blindness’ or ‘partial sight’) and ‘mortality’ (or ‘death’) and ‘prospective’ (or ‘cohort’ or ‘follow up’ or ‘follow-up’ or ‘longitudinal’). Furthermore, the reference lists of retrieved articles were reviewed for undetected relevant studies.
The included studies should fulfil the following criteria: (1) prospective studies published as original articles; (2) the participants of interest were from the general population (ie, excluding the studies in disease-specific populations, such as in patients with diabetes); (3) the exposure of interest was VI. VI is defined according to every study because the definition of VI will differ between different countries. (4) the outcome was all-cause mortality; (5) the relative risks (RRs) with corresponding 95% CIs were reported (we presented all results with RR for simplicity); (6) for dose-response analysis with restricted cubic spline models, studies must provide the number of participants or person-years, deaths and the RR with a corresponding 95% CI for each level of VA; and (7) if there was more than one study with an overlapping cohort, we just chose the latest or more complete one.
Two investigators reviewed all studies independently. If they disagreed with the eligibility of an article, they discussed it with a third investigator to resolve it.
The following information was extracted from each of the qualified studies: (1) the first author's last name; (2) publication year; (3) country where the study was performed; (4) age of objects at baseline; (5) gender of the participants; (6) follow-up duration; (7) number of participants and deaths; (8) the definition of VI and method of VI assessment; (9) variables adjusted for in the analysis; and (10) the most adjusted RR with corresponding 95% CI of the risk of mortality.
For dose-response analysis with restricted cubic spline models, the number of participants or person-years, deaths and RR with corresponding 95% CI for each level of VA were extracted. The median or mean level of VA in each level was assigned to the corresponding RR for every study. If the upper boundary of the highest level was not provided, we assumed that the boundary had the same amplitude as the adjacent level. We extracted RRs that reflected the greatest degree of control for potential confounders. There were several studies using the Snellen Notation Indicating Measurement to evaluate VA. It was converted to Logarithm of the Minimum Angle of Resolution (LogMAR) according to the Visual Acuity Measurement Standard.38
For the highest VI level versus no VI, the inverse variance-weighted mean of the logarithm of RR with a corresponding 95% CI was calculated as the pooled measure to assess the strength of associations between VI and the risk of mortality. Furthermore, the strength of association in men, women and participants older than 65 was evaluated independently. I2 was used to assess heterogeneity among studies, and I2 values of 0, 25, 50 and 75% represent no, low, moderate and high heterogeneity, respectively.39 The DerSimonian and Laird random-effect model was selected as the pooling method in this meta-analysis.40 Meta-regression with restricted maximum likelihood estimation40 and subgroup analysis were conducted to explore potential sources of heterogeneity. Publication bias was evaluated with visual inspection of the funnel plot and Egger's test.41
For a dose–response relationship, a two-stage random-effect dose–response meta-analysis was performed42 to compute the trend from the correlated log RR estimates across levels of VA. In the first stage, a restricted cubic spline model with three knots at the 10th, 50th and 90th centiles43 of the level of VA was estimated using generalised least-square regression, taking into account the correlation within each set of published RRs.43 Then the study-specific estimates were combined using the restricted maximum likelihood method in a multivariate random-effects meta-analysis.44 A p value for non-linearity was calculated by testing the null hypothesis that the coefficient of the second spline is equal to 0.
All statistical analyses were performed with STATA V.12.0 (Stata Corporation, College Station, Texas, USA). All reported probabilities (p values) were two-sided with p<0.05 being considered statistically significant.
Literature search and study characteristics
The literature search flow chart is shown in figure 1. According to our search strategy, 3008 articles were identified, of which there were 1238 articles from PubMed, 767 articles from Web of Science, 15 articles from the Chinese National Knowledge Infrastructure, 973 articles from Wanfang databases and 15 articles from reference lists. After deleting the duplications, 2421 articles were left. After reviewing the titles and abstracts, 83 articles were retrieved. Subsequently, 60 articles were excluded for various reasons after reviewing the full text. As a result, 23 articles including 29 studies involving 269 839 participants and 67 061 deaths met the inclusion criteria and were included in this meta-analysis.
Overall, 18 studies were conducted in participants older than 65 years.4 ,16 ,19 ,22 ,24 ,26–28 ,31 ,33 ,34 ,36 ,37 Eight studies were conducted in women19 ,22 ,23 ,28 ,31 ,33 ,35 ,36 and seven in men.22 ,23 ,28 ,31 ,33 ,35 ,36 With regard to the study region, nine studies were conducted in North America,4 ,16 ,18–20 ,35 ,36 four in Oceania,21 ,22 ,29 seven in Asia17 ,23 ,25 ,26 ,28 ,30 and nine in Europe.24 ,27 ,31–34 ,37 Follow-up duration ranged from 2 to 13.2 years. There were nine studies with a follow-up duration shorter than 5 years,4 ,21 ,23 ,27 ,28 ,32 ,36 six studies with a follow-up duration between 5 and 6 years,16 ,31 ,33 ,34 ,37 seven studies with a follow-up duration between 6 and 8 years17 ,22 ,26 ,30 ,35 and seven studies with a follow-up duration longer than 8 years.18–20 ,24 ,25 ,29 About VI assessment, 15 studies were assessed by acuity charts,17 ,19 ,21 ,23–27 ,29–31 ,33 ,34 three studies by the International Statistical Classification of Diseases (ICD) code,4 ,35 seven studies by self-report16 ,22 ,28 ,32 ,37 and four by other methods.18 ,20 ,36 The baseline characteristics of the studies are shown in online supplementary table S1. The detailed characteristics of the studies included in the dose–response analysis are shown in online supplementary table S2.
In the general population, 29 studies were included. Among them, 17 studies showed a significant association between VI and the risk of mortality, while the rest of the studies failed to indicate the relation between them. For the highest VI level versus no VI, the pooled RR of overall data was 1.36 (95% CI 1.25 to 1.46, I2=62.4%, Pheterogeneity<0.001, figure 2).
In the participants older than 65 years, the pooled RR of 18 studies was 1.28 (95% CI 1.18 to 1.39, I2=55.7%, Pheterogeneity=0.002, figure 3). When men and women were analysed separately, the pooled RR of the risk of mortality was 1.29 (95% CI 1.07 to 1.54, I2=60.4%, Pheterogeneity=0.019) in men and 1.39 (95% CI 1.14 to 1.70, I2=74.0%, Pheterogeneity<0.001) in women.
Dose-response analysis with restricted cubic spline functions
For dose–response analysis between VA and the risk of mortality, six studies including 10 922 deaths19 ,23 ,25 ,27 ,30 were included. According to ICD-10, VA equal to or better than 0.48 LogMAR was considered as mild or no VI, and VA worse than 0.48 LogMAR was VI.1 A linear relation was found between VA and the risk of mortality (Pnon-linearity=0.38). Compared with 0 LogMAR, the risk of mortality increased significantly with increasing LogMAR and the RRs (95% CIs) of mortality were 1.03 (1.01 to 1.06), 1.09 (1.04 to 1.15), 1.23 (1.15 to 1.31) and 1.31 (1.11 to 1.55) for 0.09, 0.25, 0.75 and 1.20 LogMAR, respectively. The dose–response analysis suggested that the risk of mortality increased by 4% (RR: 1.04, 95% CI 1.01 to 1.06) for a 0.1 Log MAR increment of VA (figure 4).
Meta-regression and subgroup analyses
As seen in figure 2, moderate heterogeneity (I2=62.4%, Pheterogeneity<0.001) among all included studies was demonstrated. p Values of meta-regression analysis with the covariates of continent, publication year, VI assessment and follow-up duration were 0.010, 0.288, 0.002 and 0.004, respectively. The results indicated that the covariates of continent, VI assessment and follow-up duration were sources of between-study heterogeneity. In view of the results of meta-regression analysis, subgroup analyses were performed by continent, VI assessment and follow-up duration and all subgroup analysis manifested that VI was associated with a statistically significant increased risk of mortality. For instance, when the analysis was restricted to the studies that were, respectively, conducted in four continents, the pooled RRs were 1.29 (95% CI 1.16 to 1.44) in North America, 1.44 (95% CI 1.26 to 1.65) in Oceania, 1.80 (95% CI 1.42 to 2.26) in Asia and1.22 (95% CI 1.03 to 1.43) in Europe (the details are shown in table 1).
Influence analysis and publication bias
No individual study had an excessive influence on the aforementioned pooled effects in influence analysis. The Egger's test and funnel plot detected no evidence of significant publication bias for the analysis of all included studies (p=0.110; figure 5), for studies in participants older than 65 years (p=0.803), studies in women (p=0.258) and studies in men (p=0.794).
To our knowledge, this is the first meta-analysis to evaluate the association between VI and the risk of mortality. The results demonstrated that VI was significantly associated with an increased risk of mortality. Significant associations were observed in participants older than 65 years, in men and women and in other stratified analysis. There was a notable linear relationship between VA and the risk of mortality, and the risk of mortality increased by 4% for a 0.1 LogMAR increment of VA.
For the relationship between VI and the risk of mortality, the mechanism is not clear now. Nevertheless, there are some possible explanations to justify this. (1) VA could reflect the functional status and VI would give rise to numerous functional problems, such as falls,45 accidents,9 loss of independence10 ,46–48 and depression,49 which may be life-threatening. (2) VA is associated with signs of frailty, and frailty predicts mortality.50 (3) Visual system is part of the central nervous system. With increasing age, the number of retinal ganglion cells will decrease in the visual system. Therefore, it is easy to understand that VI, as a reflection of biological ageing, will increase the risk of mortality.51 (4) It is possible that there are common underlying genetic mechanisms between eye diseases and the risk of mortality.51 All the above factors could increase the risk of mortality with VI.
Between-study heterogeneity is common and exploration of potential sources of between-study heterogeneity is essential. Our meta-analysis showed moderate between-study heterogeneity. To explore the sources of the between-study heterogeneity, the meta-regression was carried out with the covariates of continent, publication year, VI assessment and follow-up duration. The covariates of continent, follow-up duration and VI assessment were found to influence between-study heterogeneity (p<0.05). When the analysis was restricted to the studies that were respectively conducted in four continents, the heterogeneity decreased in Oceania and Europe. Subgroup analysis was conducted by follow-up duration, and the heterogeneity decreased in subgroups of follow-up duration between 6 and 8 years and a follow-up duration longer than 8 years. Regarding VI assessment, the results of subgroup analysis showed that the heterogeneity decreased in subgroups by self-report and by other methods. Above all, the results of subgroup analysis by continent, follow-up duration and VI assessment remained stable. Some original studies did not adjust covariates related to mortality (age, gender, living alone etc.), which may lead to the instability of results. Therefore, the different categories and quantities of adjusted covariates in the original studies may be a source of between-study heterogeneity. What's more, the reason for the moderate level of heterogeneity may be due to clinical differences in diagnosis of VI between studies.
There are several strengths in our study. First of all, a major advantage of this meta-analysis is that all the included studies are prospective studies; thus, the significant association between VI and the risk of mortality could indicate a potential causal relationship between them. Second, a dose–response analysis was performed to quantitatively describe the association between them. Third, this meta-analysis contained a large number of participants, allowing us to report credible results. Fourth, all results at the highest VI level versus no VI and subgroup analysis were consistent, which indicated that our results were stable and reliable.
There are also several limitations in our study. First, the methods of VI assessment varied. Fifteen studies were assessed by acuity charts,17 ,19 ,21 ,23–27 ,29–31 ,33 ,34 and the acuity charts contained many kinds of charts (Snellen E chart, Bailey-Lovie letter chart, Landolt ring chart, etc) and three studies by ICD code,4 ,35 seven studies by self-report16 ,22 ,28 ,32 ,37 and four by other methods.18 ,20 ,36 The different methods may lead to a misclassification of objects. Then the misclassification may have an impact on the results. Second, although we extracted RRs (95% CI) that reflected the greatest degree of control for potential confounders, the categories and quantities of the adjusted covariate varied in the original studies. For example, one study adjusted for many covariates, such as age, living alone, stroke, smoking, etc.36 However, another study did not adjust for the covariates of age, living alone and smoking.22 Third, all the included studies were prospective cohort studies, but the different follow-up duration (2–13.2 years) may have a potential impact on the results.
Taken together, the principal finding of this meta-analysis was that persons with VI had a higher risk of mortality than persons without VI. Our findings further confirmed that decreased vision may increase the risk of mortality. Therefore, the association between VI and the risk of mortality could be mediated by increasing the risks of accidents9 and the need for community services10 and institutionalisation.11 ,12 All the adverse health effects may increase the risk of mortality, especially in the ageing population. Further research is needed to explore why decreased vision is associated with a higher risk of mortality.
What is already known on this subject
Many studies found that visual impairment impacted people's health by increasing the risks of falls, fractures, accidents, etc.
Many studies have been conducted to assess the association of visual impairment with the risk of mortality and the results remain controversial.
What this study adds
This is the first meta-analysis to evaluate the association between visual impairment and the risk of mortality.
This meta-analysis shows that visual impairment was significantly associated with an increased risk of mortality.
Contributors TZ and WJ conceived of the study, participated in its design and coordination, carried out the literature search and data extraction, and were involved in the drafting of the manuscript or its critical revision for important intellectual content. XS carried out the literature search, data extraction and interpretation of the data. DZ was involved in the interpretation of the data and in the critical revision of the manuscript for important intellectual content.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.